Study design and data source. The study used data from the Global School–based Student Health Survey (GSHS). The GSHS is a large a cross–sectional study that collects a range of health data including the risk and protective factors associated with adolescent initiation of alcohol, drug, and tobacco use. It is a collaborative surveillance project designed to help countries measure and assess the behavioural risk and protective factors in ten key areas among young people aged 14–17 years. The GSHS is a low–cost school–based survey which uses a self–administered questionnaire to obtain data on young people’s health behaviour and protective factors related to the leading causes of morbidity and mortality among adolescents aged 14–17 years. The GSHS provides an excellent opportunity to investigate patterns in early exposure to risk factors across LMICs. The data are publicly available22. The primary aim of the survey is to inform the development of evidence–based adolescent and school health policies and programmes in countries22,23. The survey uses the same standardised sampling techniques and study methodology across countries. The questions in each GSHS are tailored to each country context but the study design and participants’ section procedures are similar across the 101 GSHS countries22. Questionnaire items included questions on demographics (e.g., age, sex), food insecurity, peer victimisation, loneliness, and anxiety, peer conflict and injured (physically attacked, fighting and seriously injured), health hazard initiations (e.g., drinking alcohol, drug and tobacco use), parental control (parents check homework, parents understand problem, parent monitoring), lifestyle factors (leisure–time sedentary behaviours, physical activity). For countries with more than one GSHS dataset, we used the most recent one available. A total of 303,521 participants from 92 countries were included in our analysis (Appendix Table A1). The analytical sample consists of participants aged 14–17 years from 92 countries, including 15 low–income, 33 lower–middle income, 29 upper middle–income and 15 high income countries based on the World Bank classification24 at the time of the survey.
Measures
Outcome variables. We used two outcome variables: adolescence health hazard early initiations (i.e., drinking alcohol, drug and tobacco use) and the number of health hazard initiations (Appendix Table A2). Each of these outcome variables was measured with a single self–reported item or question. By extending the analytical explorations, number of reported health hazard initiations among the adolescents were also considered an outcome variable. Response were categorised as ‘none’ if the participants reported that they had not experienced any form of health hazards initiation or as ‘one health hazard initiation’’ if they reported having experienced at least one health hazard initiation; ‘two health hazard initiations’ if they reported having experienced two health hazard initiations; or ‘three health hazard initiations’ if they reported that they had experienced three health hazard initiations (Appendix Table A2).
Explanatory variables
Violence and unintentional injury related factors. Violence and unintentional injury were assessed with the questions in the measure (how often students have been physically attacked, how often they have participated in a physical fight, frequency of serious injuries, and frequency of bullying). Physical violence by peers was assessed with the questions: “During the past 12 months, how many times you were physically attacked” and “During the past 12 months, how many times were you in a physical fight?”. Student responses for being physically attacked and fighting were recoded as ‘yes’ (reported being attacked or fighting one or more times) or ‘no’ otherwise. The status of student’s serious injuries was defined as ‘yes’ if they reported being seriously injured one or more time according to the question “During the past 12 months, how many times were you seriously injured?” or ‘no’ otherwise. Participants’ bullying victimisation was defined as dichotomised (1 = 'yes' if the participant reported bullying experiences on one or more days, or 0 = 'no' otherwise).
Psychological factors. Two psychological factors included in this study were anxiety and loneliness. Participant’s level of anxiety and loneliness were measured using the following questions: ‘During the past 12 months, how often have you been so worried about something that you could not sleep at night?’ and ‘During the past 12 months, how often have you felt lonely?’. These responses were coded as never, rarely or sometimes, most of the time, or always.
Protective factors. Protective factors were added in terms of measuring peer’s social support at school and parental regulation and monitoring. Parental regulation and monitoring were assessed as the role of parental support using three variables of parents checking homework, parents understanding the problem, and parental monitoring. Responses were recorded as never, rarely or sometimes, most of the time, or always (Appendix Table A2).
Lifestyle factors. Lifestyle factors included questions on food insecurity, sedentary behaviours and obesity. Respondent food insecurity was measured and responses were recoded ‘most of the time or always’ as ‘severe food insecurity (Q1)’, ‘rarely or sometimes’ as ‘moderate food insecurity (Q2)’, and ‘never’ as ‘food security (Q3)’. There were questions that asked participants about time spent engaged in sitting activities and watching television as well as their weight and height. Students were asked: ‘How much time do you spend during a typical or usual day sitting and watching television, playing computer games, talking with friends, or doing other sitting activities. the following question: Student’s everyday sitting activities were categorised as follows: none, <1 hour/day, 1–2 hours/day, 3–4 hours/day, and ≥ 5 hours/day (Appendix Table A2).
Statistical analyses. Weighted estimates of prevalence were expressed with corresponding 95% confidence intervals (CIs) for the national and regional perspectives. The data were weighted by sampling weight to make the estimate nationally representative. All analyses were weighted using sampling unit (PSU) which is derived from the probability of a school being selected, a classroom being selected, school and student level non–response and gender. We conducted multinomial logistic regression analyses to examine the factors associated with health hazard initiation (drinking alcohol, drug, and tobacco use) and magnitude of health hazard initiation. We conducted unadjusted multinomial logistic regression to select variables using only separated explanatory variables, which had a bivariate association with health hazard early initiations and magnitude of health hazard initiations. In the analytical exploration, statistical significance was considered at the 5% risk level. All analyses were performed using the statistical software Stata/SE 13 (StataCorp, College Station, Texas, USA).
Ethics consideration. The study used secondary data from the Global School-based Student Health Survey (GSHS), which had already been ethically approved by the respective participating countries and the dataset is deidentified; as a result, we did not require any further ethical approval. Details of the ethical procedures followed by the GSHS Program can be found in their weibsite (https://www.cdc.gov/gshs/index.htm).